AI Transaction Pattern Analysis Integration

I’m building a new payment-processing platform and need an AI module that zeroes in on transaction analysis, with a strong emphasis on pattern recognition. The goal is to surface reliable insights from live and historical payment streams so we can flag outliers, enrich customer dashboards, and inform our risk-scoring engine.

I need someone who can move fast—ASAP delivery is essential—yet still build something maintainable. You’re free to choose the most suitable stack (Python, Scala, or another language you can justify) and libraries such as PyTorch, TensorFlow, scikit-learn, or Spark ML; just be prepared to explain why your choices fit a high-volume payments environment.

Please base the model on real-world transactional datasets (synthetic augmentation is fine) and include a clear path for retraining as new data arrives. All code should be production-ready, containerised, and exposed through a REST or gRPC endpoint that my back-end can call.

Deliverables • Data preprocessing scripts and reproducible training pipeline • Pattern-recognition model with inference service (Docker/K8s ready) • Unit and load tests proving stability at 1k TPS+ • Minimal latency benchmarks and optimisation notes • Setup & usage documentation (Markdown)

Acceptance criteria • End-to-end demo on sample payment feed showing correct pattern extraction • CI pipeline passes all tests in our staging environment • Response time ≤200 ms for single inference call on commodity hardware

If you have prior experience embedding ML inside payment or fintech stacks, I’d love to see it—code samples, repos, or short case studies are welcome. Budget: USD 699–700 Skills: Java, Python, Data Processing, CUDA, Machine Learning (ML), Data Analysis, Data Integration, REST API